Method for regulating and controlling evolution trend of acoustic emission analog signal

Abstract

The invention discloses a method for regulating and controlling the evolution trend of an acoustic emission analog signal, and relates to the field of nondestructive testing of composite materials. The method comprises the following steps: firstly, acquiring an actual acoustic emission signal, and screening out a signal of a specific damage mode through fast Fourier transform (FFT); carrying out wavelet packet noise reduction processing on the collected acoustic emission signals, and extracting representative acoustic emission signals as target signals by using a principal component analysis (PCA) method; then, a BP neural network model is constructed for training, and an acoustic emission signal envelope line is predicted; selecting an optimal fitting model to perform envelope line fitting by comparing R square values; determining a sudden change point through a threshold value and a sliding window technology, and intercepting a specific curve segment of the optimal fitting model; and finally, selecting an optimal loading evolution trend regulation and control time point through weighted evaluation of a root-mean-square error (RMSE) and a characteristic frequency change value. Aiming at the problem that the acoustic emission signal simulated by a finite element does not conform to an actual signal, the invention provides the acoustic emission signal evolution trend regulation and control method, the analog signal is corrected, the accuracy is effectively improved, and trend matching of the acoustic emission analog signal of the composite material and the actual signal is realized.

Description

Technical Field

[0001] The invention belongs to the field of acoustic emission nondestructive testing, and in particular relates to a method for regulating the evolution trend of an acoustic emission simulation signal. Background Art

[0002] Glass fiber reinforced composites (GFRP) have been widely used in key fields such as aerospace, petrochemicals, and automobiles due to their high specific strength and fatigue corrosion resistance. During use, composite materials are subjected to various complex stresses and environmental factors for a long time, and are prone to damage, such as matrix cracking, fiber breakage, and delamination. If these damages are not discovered and handled in time, they may lead to serious accidents such as blade breakage and container rupture, which will not only cause huge economic losses, but also endanger the lives of personnel. If the damage can be monitored and assessed in time, a maintenance or replacement plan can be formulated in time to reduce economic losses. Among the various existing detection methods, acoustic emission technology has good application prospects in structural health monitoring due to its advantages of dynamic monitoring, high sensitivity, and strong real-time performance.

[0003] In actual processes, the characteristic frequency bands of acoustic emission signals generated by different materials and different types of damage are different, so for each material, a large number of damage experiments need to be carried out in advance to determine the various types of damage characteristics, which is not only time-consuming and labor-intensive, but also inefficient. Obtaining acoustic emission signals with different damage characteristics through finite element simulation can greatly reduce work intensity and improve work efficiency.

[0004] However, finite element simulation cannot directly add attenuation changes when acquiring acoustic emission signals, which leads to significant differences between simulated signals and actual signals. The acoustic emission signals have no attenuation and show a trend of infinitely increasing amplitude, which seriously deviates from the actual monitoring situation, thus affecting the accuracy and reliability of monitoring. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an evolution trend control method of an acoustic emission simulation signal, which is used to correct the acoustic emission simulation signal to achieve trend matching with the actual acoustic emission signal, specifically:

[0006] Step 1: Obtain the actual acoustic emission signal, filter out the signal of the specific damage mode through fast Fourier transform (FFT), and perform wavelet packet noise reduction to remove noise interference and improve signal quality;

[0007] Step 2: Use principal component analysis (PCA) to obtain the target signal, build a BP neural network model, train the neural network, and predict the envelope of the acoustic emission signal;

[0008] Step 3: Fit the predicted envelope and ensure the effectiveness of the fitting model by comparing the R-squared value;

[0009] Step 4: Use threshold and sliding window technology to detect the mutation points of the simulated signal and the actual signal;

[0010] Step 5: Match the sampling rates of the simulated signal and the actual acoustic emission signal, apply the fitting model, and intercept the attenuation curve segment for subsequent evolution trend regulation;

[0011] Step 6: A comprehensive evaluation method using weighted evaluation of the root mean square error (RMSE) between the trend-controlled signal and the actual signal and the change in characteristic frequency before and after processing is used to determine the starting time of the evolution trend control and to control the evolution trend of the acoustic emission simulation signal with infinite increase characteristics.

[0012] Compared with the prior art, the present invention aims at the problem that the acoustic emission signals obtained by the current finite element simulation cannot take into account the signal attenuation, resulting in serious discrepancies between the simulated signal and the actual signal, and proposes a method for regulating the evolution trend of acoustic emission simulation signals. A BP neural network is constructed to predict the signal envelope, fit the envelope and intercept the attenuation curve segment. Combined with the weighted evaluation of the root mean square error and the characteristic frequency change value, the optimal loading evolution trend regulation time point is automatically selected to achieve the evolution trend regulation of the acoustic emission simulation signal. The acoustic emission signal evolution trend regulation method provided by the present invention is effective and feasible, and has a high signal matching degree. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] By describing in detail the embodiments shown in conjunction with the accompanying drawings, the drawings described below are only some embodiments of the present invention, in which:

[0014] Figure 1 It is an overall flow chart of an evolution trend control method of an acoustic emission simulation signal;

[0015] Figure 2 This is the training graph of the constructed BP neural network;

[0016] Figure 3 It is the regression coefficient diagram of BP neural network;

[0017] Figure 4 Normalized envelope diagram: (a) actual acoustic emission signal envelope; (b) predicted acoustic emission signal envelope

[0018] Figure 5 It is a comparison chart of different fitting models and prediction envelopes;

[0019] Figure 6 It is a bar chart of R-squared values ​​of different fitting models;

[0020] Figure 7This is the simulated signal diagram of acoustic emission for matrix damage;

[0021] Figure 8 This is the acoustic emission signal diagram of matrix damage obtained in the experiment;

[0022] Fig. 9 This is the simulated signal diagram of acoustic emission of matrix damage after evolution trend regulation;

[0023] Fig.10 The spectrum diagram of the acoustic emission simulation signal before and after processing; DETAILED DESCRIPTION

[0024] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. The present invention proposes a method for controlling the evolution trend of an acoustic emission simulation signal. Figure 1 This is the flow chart of the method. After obtaining the actual acoustic emission signal, the fast Fourier transform (FFT) is used to filter out the signal of the specific damage mode. After wavelet packet denoising, the principal component analysis (PCA) is used to extract the target signal, and the BP neural network prediction envelope is constructed and fitted. The mutation point is determined by the threshold and sliding window technology, and the attenuation curve segment of the optimal fitting model is intercepted. The root mean square error and characteristic frequency weighted evaluation are used to determine the time starting point of the evolution trend regulation, and the evolution trend of the acoustic emission simulation signal is regulated to achieve trend matching with the actual acoustic emission signal.

[0025] The design and production scheme of the composite pressure vessel laminate [45°] ply specimen prefabricated in this embodiment refers to GB / T1446-2005 General Rules for Test Methods for Fiber Reinforced Plastics. The specimen is a 45° ply with 7 layers, and the specific dimensions are 100 mm in length, 100 mm in width, and 5 mm in thickness.

[0026] The prepared laminate [45°] ply specimen was installed on the universal testing machine, and the support roller length was set to 40mm. An acoustic emission sensor was arranged on the upper edge of the specimen to collect the acoustic emission signal of the entire bending process. The acoustic emission sensor was 40mm away from the center. The acoustic emission sensor was coupled and fixed with a special coupling agent. After that, the acoustic emission sensor was connected to the amplifier. The amplifier parameters were designed to be 40dB, and the acoustic emission acquisition parameters were adjusted as shown in Table 1. Table 1 Acoustic emission acquisition parameter setting values Parameter name Parameter Value Threshold 45dB Preamplifier gain 45dB Sampling rate 1MSPS(1MHz) Filter range 1KHz~500KHz

[0027] After the acoustic emission equipment is prepared, the bending loading is then started at a loading speed of 1mm / min. During the loading, the acoustic emission sensor is used to monitor the acoustic emission signals generated during the entire stretching process. The stretching machine stops working until the sample breaks.

[0028] All acoustic emission signals obtained in the experiment were processed by fast Fourier transform (FFT) to obtain acoustic emission signals with characteristic frequencies falling between 20kHz and 60kHz (i.e., acoustic emission signals of matrix damage), and a data set for BP neural network was obtained.

[0029] Select dB5 as the wavelet basis function and decompose it into 4 layers, and then use the general threshold method to calculate the threshold: Where σ is the standard deviation of noise, N is the signal length, and then the coefficients less than the threshold are set to zero to reduce noise. Finally, the retained significant coefficients are used for inverse transformation to obtain the denoised signal.

[0030] Organize the obtained multiple groups of denoised acoustic emission signals into a matrix X, where each row represents a sample and each column represents a time point, and centralize the data. Calculate the covariance matrix: Then, the eigenvalues ​​and eigenvectors of the covariance matrix are solved to obtain the projection values ​​of all signals on the first principal component, and the signal with the largest projection in the direction of the first principal component is selected as the target signal.

[0031] The denoised acoustic emission signal is subjected to Hilbert transform to obtain the analytical signal, and then the absolute value of the analytical signal is taken to obtain the envelope, and then the envelope amplitude is normalized.

[0032] Set the number of hidden layers to 18, and then set the transfer function. The hidden layer uses the tansig hyperbolic tangent S-type function, the output layer uses the purelin linear function, and the Levenberg-Marquardt (trainlm) algorithm is used. 70% is divided into a training set, 15% is a validation set, and 15% is a test set. The parameter settings are as follows: Figure 2 As shown in the figure, the maximum number of training rounds is set to 500, the maximum number of verification failures is set to 6, and the initial value of the learning rate is 0.01. The entire training process is that the input data is the preprocessed envelope, the target output is the target signal selected by the principal component analysis (PCA), the network weights are adjusted by back propagation, and the predicted envelope is obtained by using the trained network for prediction. Finally, Figure 3 The regression coefficient shown in the figure evaluates the training effect. If the regression coefficient is lower than 0.9, retraining is performed. The predicted results after training are as follows Figure 4 shown.

[0033] Fitting is performed on the obtained prediction envelope, including polynomial fitting, sine fitting, Fourier fitting, Gaussian fitting and rational function fitting, such as Figure 5 As shown, the polynomial fitting tried polynomials of order 1-10, and the R square values ​​of these five fitting models were compared, as shown in Figure 6 As shown, the best fitting model is selected.

[0034] The signal mutation point is detected by threshold and sliding window technology, and the acoustic emission simulation signal is imported ( Figure 7 ) and actual acoustic emission signal data ( Figure 8 ), the smoothed analog signal and actual signal are obtained by moving average filtering and low-pass filtering. The dynamic threshold is calculated by multiplying the standard deviation of the smoothed signal by the set amplitude mutation threshold. The change of the signal is calculated, and the absolute value array of the change is traversed by the sliding window method. The first point greater than the dynamic threshold is the mutation point. If the analog signal does not detect the mutation point, fault tolerance processing is performed, the dynamic threshold is halved, and the mutation point is re-detected.

[0035] Calculate the ratio of the actual signal mutation point position to the total length of the signal, and extract the corresponding data segment from the simulated signal. Perform linear interpolation resampling on the actual signal to make it correspond to the sampling rate of the simulated signal. Extract the normalized fitting model from the maximum value to the end point of the model as the attenuation curve segment for subsequent evolution trend regulation.

[0036] The weight of the root mean square error (RMSE) is set to 0.7, and the weight of the frequency change is set to 0.3. Try to load the attenuation curve segment at different positions of the simulated signal for evolution trend control. After each application, calculate the RMSE of the actual signal and the frequency characteristic changes before and after processing. Combine these two indicators through weights to form a comprehensive evaluation score. The lower the score, the better the control effect. After traversing all possible starting positions, select the position with the lowest score as the optimal time starting point for evolution trend control. The evolution control results are shown in Figure 2. Fig. 9 As shown in the figure, the influence of evolution trend regulation on the original acoustic emission simulation signal is Fig.10 It can be determined from the figure that the evolution trend control will not change the characteristic frequency of the acoustic emission signal and will not affect the characteristic identification of the damage mode.

[0037] Finally, it is necessary to explain here that the above-mentioned specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Some non-essential improvements and adjustments made by technicians in this field based on the above-mentioned contents of the present invention belong to the protection scope of the present invention.

Claims

1. A method for controlling the evolution trend of an acoustic emission simulation signal, characterized in that: The method comprises the following steps: Step 1: Obtain the actual acoustic emission signal, filter out the signal of the specific damage mode through fast Fourier transform (FFT), and perform wavelet packet noise reduction to remove noise interference and improve signal quality; Step 2: Use principal component analysis (PCA) to obtain the target signal, build a BP neural network model, train the neural network, and predict the envelope of the acoustic emission signal; Step 3: Fit the predicted envelope and ensure the effectiveness of the fitting model by comparing the R-squared value; Step 4: Use threshold and sliding window technology to detect the mutation points of the simulated signal and the actual signal; Step 5: Match the sampling rates of the simulated and actual acoustic emission signals, apply the fitting model, and intercept the attenuation curve segment for subsequent evolution trend regulation; Step 6: A comprehensive evaluation method using weighted evaluation of the root mean square error (RMSE) between the trend-controlled signal and the actual signal and the change in characteristic frequency before and after processing is used to determine the starting time of the evolution trend control and to control the evolution trend of the acoustic emission simulation signal with infinite increase characteristics.

2. The method for controlling the evolution trend of an acoustic emission simulation signal according to claim 1, characterized in that: In step 1, the actual acoustic emission signal is obtained, and the signal of the specific damage mode is screened out by fast Fourier transform (FFT), and wavelet packet noise reduction processing is performed to remove noise interference and improve signal quality. The specific steps are as follows: Step 1.1: Perform a three-point bending test on the composite laminate and use an acoustic emission sensor to collect acoustic emission signals during the loading process; Step 1.2: Perform fast Fourier transform (FFT) on the collected acoustic emission signals to extract the acoustic emission signals of the characteristic frequencies of specific damage modes; Step 1.3: Perform wavelet packet denoising on all collected acoustic emission signals, set dB5 as the wavelet basis function and decompose it into 4 layers, and use the general threshold method to calculate the threshold: Where σ represents the standard deviation of noise and N is the signal length. Then the coefficients less than the threshold are set to zero to complete the noise reduction operation, and finally the retained significant coefficients are used for inverse transformation to obtain the noise-reduced signal.

3. The method for controlling the evolution trend of an acoustic emission simulation signal according to claim 1, characterized in that: In step 2, the target signal is obtained by using principal component analysis (PCA), a BP neural network model is constructed, the neural network is trained, and the envelope of the acoustic emission signal is predicted. The specific steps are as follows: Step 2.1: Arrange the denoised acoustic emission signals into a matrix X, where each row corresponds to a sample and each column corresponds to a time point. After centering the data, calculate the covariance matrix: Afterwards, the eigenvalues ​​and eigenvectors of the covariance matrix are solved to obtain the projection values ​​of all signals on the first principal component, and the signal with the largest projection in the direction of the first principal component is selected as the target signal; Step 2.2: Use Hilbert transform to transform the denoised acoustic emission signal to obtain the analytical signal, then take the absolute value of the analytical signal to obtain the envelope, and normalize the envelope amplitude; Step 2.3: Construct a BP neural network, determine the number of hidden layer nodes to be 18, configure the transfer function, where the hidden layer uses the tansig hyperbolic tangent S-type function, and the output layer uses the purelin linear function. Use the Levenberg-Marquardt (trainlm) algorithm for training, divide the signal into 70% training set, 15% validation set and 15% test set, set the maximum number of training rounds to 500, the maximum number of validation failures to 6, and the initial value of the learning rate to 0.01; Step 2.4: Take the envelope as input and the target signal selected by principal component analysis (PCA) as output, adjust the network weights through the back propagation algorithm, train the network and obtain the predicted envelope.

4. The method for controlling the evolution trend of an acoustic emission simulation signal according to claim 1, characterized in that: In step 3, the predicted envelope is fitted, and the effect of the fitting model is ensured by comparing the R square value. The specific steps are as follows: Step 3.1: Based on the predicted envelope obtained in step 2, polynomial fitting (1st to 10th order), sine fitting, Fourier fitting, Gaussian fitting and rational function fitting are used respectively; Step 3.2: Compare and analyze the R-squared values ​​of the fitted models and select the best fitting model.

5. The method for controlling the evolution trend of an acoustic emission simulation signal according to claim 1, characterized in that: In step 4, the threshold and sliding window technology are used to detect the mutation points of the simulated signal and the actual signal. The specific steps are as follows: Step 4.1: Obtain the smoothed analog signal and actual signal by moving average filtering and low-pass filtering, and calculate the dynamic threshold by multiplying the standard deviation of the smoothed signal by the set amplitude mutation threshold; Step 4.2: Calculate the change of the signal and use the sliding window method to traverse the absolute value array of the change. The first point that is greater than the dynamic threshold is the mutation point. If the simulation signal does not detect the mutation point, perform fault tolerance processing, halve the dynamic threshold, and re-detect the mutation point.

6. The method for controlling the evolution trend of an acoustic emission simulation signal according to claim 1, characterized in that: In step 5, the sampling rates of the simulated signal and the actual acoustic emission signal are matched, the fitting model is applied, and the attenuation curve segment is intercepted for subsequent evolution trend regulation. The specific steps are as follows: Step 5.1: Calculate the ratio of the actual signal mutation point position to the total signal length, and extract the corresponding data segment from the simulated signal; Step 5.2: Perform linear interpolation sampling on the actual signal to match the sampling rate of the analog signal; Step 5.3: Perform amplitude normalization on the simulated signal and the actual signal. In the normalized fitting model, the interval from the maximum value point to the end point of the model is intercepted as the attenuation curve segment for subsequent evolution trend regulation.

7. The method for controlling the evolution trend of acoustic emission signal simulation according to claim 1, characterized in that: In step 6, a comprehensive evaluation method of weighted evaluation of the root mean square error (RMSE) between the trend-controlled signal and the actual signal and the change value of the characteristic frequency before and after processing is adopted to determine the starting time of the evolution trend control, and to control the evolution trend of the acoustic emission simulation signal with infinite increase characteristics. The specific steps are as follows: Step,6.1: Set the weight of root mean square error (RMSE) to 0.7 and the weight of frequency change to 0.3; Step 6.2: Use the attenuation curve segment to adjust the evolution trend from different positions of the simulated signal, calculate the RMSE after each adjustment and the change in frequency characteristics before and after processing, and combine these two indicators according to the weight to form a comprehensive evaluation score. The lower the score, the better the adjustment effect; Step 6.3: After traversing all possible starting positions, select the position with the lowest score as the optimal evolution trend control time starting point to control the simulation signal of the infinitely increasing feature.

Images